Dissertation Defense | Mar 19, 2pm | Ferhat Erata (Yale) - Learning Randomized Reductions and Program Properties for Security, Privacy, and Side-Channel Resilience

[Aviv Yaish]

Hi all,

All are invited to join the dissertation defense of Ferhat Erata, which will start in 1 hour.

• When: Wednesday (Mar 19), 2pm

• Where: DL220, Dunham Lab, 10 Hillhouse Ave, New Haven, CT 06511, US

• Speaker: Ferhat Erata (Yale)

• Title: Learning Randomized Reductions and Program Properties for Security, Privacy, and Side-Channel Resilience (PhD dissertation defense)

• Abstract: Modern computing systems face multifaceted challenges in security, privacy, and leakage resilience. This dissertation makes four key contributions to addressing these challenges. First, it focuses on analyzing side-channel vulnerabilities in low-level cryptographic code and quantum computers using symbolic AI techniques. I introduce novel symbolic register analyses to automatically detect power side-channel vulnerabilities in constant-time cryptographic implementations. Additionally, I demonstrate an algebraic reconstruction method to reverse-engineer quantum circuits from power traces, aiming to extract proprietary information from these circuits. Second, my research explores learning randomized reductions. Informally, a randomized self-reduction allows computing a function’s value at a specific point by evaluating it on randomized inputs. Here, I present a new framework that dynamically infers such properties from implementations using machine learning. Third, the dissertation demonstrates practical applications of these randomized reductions in compiling effective countermeasures against power side-channel and fault injection attacks. It also develops protocols for leakage-resilient machine learning and private quantum computations. Finally, it investigates learning-based methods for partitioning propositional encodings of combinatorial security analysis problems within the cube-and-conquer paradigm, which splits large SAT instances into smaller, more tractable subproblems. We train transformer models to learn branching heuristics within SAT-solving frameworks. Together, these contributions advance automated security analysis and resilience across classical and quantum domains.
  Advisors: Ruzica Piskac, Jakub Szefer (co-advisor). Committee: Zhong Shao, Shafi Goldwasser (UC Berkeley), Byron Cook (AWS), Scott Shapiro.

• Livestream: https://yale.zoom.us/my/ferhat

• More: For additional details about the talk and our seminar, see our website: https://yacl.cs.yale.edu